An e-commerce chatbot does two things well: answering frequent questions and tracking an order, at any hour. It does a third thing badly: advising in order to sell. That confusion explains why so many customers can't stand chatbots — in France, Ipsos BVA ranks them the least-liked channel of all. People expect advice and get an answer beside the question, or a loop with no exit. So the question to ask is not “chatbot or no chatbot”, but “support or selling?”. The gap is measured: 154% more conversion among visitors who have a conversation than among those who don't (Gorgias, 2026). This guide gives you the criteria to choose without being sold hot air.
What an e-commerce chatbot genuinely does well
An e-commerce chatbot is very effective at three precise tasks: answering frequent questions (opening hours, delivery, returns), tracking an order in progress, and routing someone to the right page or the right team. Within that scope it works: the customer gets an immediate answer, at night and at weekends, without waiting for an email or a phone call.
It is a real saving for the team, who no longer repeat the same answers all day. That part is not the problem. The problem starts when the same tool is expected to advise a customer hesitating between two products, or to reassure them about a choice. That is where most general-purpose chatbots come off the rails, because it was never their job: they were built to classify requests, not to know a catalogue.
Why so many customers can't stand chatbots
Customers don't like chatbots. In France, Ipsos BVA (5,000 people surveyed, 2025) ranks them the least-liked channel of all, well behind the phone or email. Three reasons come up every time. First, answers beside the question: the customer asks for advice and gets a link to a generic page. Second, loops with no exit: a menu that always returns to the same starting point, never reaching a human when one is needed. Third, no real advice: the chatbot can say « here is our returns policy », but not « this product suits your situation ».
The risk is not neutral. One bad experience is enough to lose a customer who actually liked the brand: 32% of consumers walk away from a brand they love after a single bad experience (PwC, 15,000 consumers surveyed). A badly calibrated chatbot can therefore cost more than the time it saves.
The real question: support, or selling?
The real question to ask before choosing a chatbot is this: are you trying to close tickets, or to prepare sales? Those are two different jobs with two different logics. One closes the request, the other leads to the cart — the difference between answering a question and accompanying a buying decision.
That distinction is not a marketing detail; it is measured. Gorgias's State of Conversational Commerce 2026 (16,000 brands analysed) shows a 154% conversion gap between visitors who have a conversation and those who don't. At bareMinerals, a cosmetics brand, this kind of conversational guidance returned 8.8 times the investment in the first month (Gorgias case study). At Topicals, a skincare brand, sales driven by customer advice grew 78% (case study published by Gorgias). None of those figures describe a classic support chatbot: they measure the effect of a conversation that advises, not one that redirects. To dig into the difference between the two categories of tool, our page on AI agents versus chatbots sets out the technical criteria.
The selection criteria that actually matter
Five criteria separate a tool that will genuinely help your sales from one that merely occupies the chat window. First: real knowledge of the catalogue. A tool that only knows generic scripts will never recommend one precise product from yours.
Second: bounded reliability. On subjects that carry weight — an allergy, a pregnancy, a complaint — the tool has to know when to stop and hand over, rather than invent a reassuring but wrong answer. Third: a real human escalation, not a contact form buried in a sub-menu. Fourth: figures, but attributed. Be wary of any « +40% conversion » claim with no source or method cited: a serious figure always comes with its study, its sample and its date. Fifth: unified channels. A customer who starts on the site and finishes on WhatsApp must find the same memory of the conversation, not start again from zero. Our page on deploying an AI on a Shopify store sets out how to check these criteria concretely before installing.
When a general-purpose chatbot is enough, and when it costs you sales
A general-purpose chatbot is perfectly sufficient when your need stops at support: opening hours, order tracking, returns policy, logistics FAQ. For that scope, a simple, inexpensive tool does the job — there is no point paying more.
It costs you sales, however, as soon as your products need real advice to be bought with confidence: technical products, broad ranges with many possible choices, products that touch health or appearance. In those cases a hesitating customer who finds nobody to advise them often leaves without buying, or buys elsewhere. That is exactly the gap the 154% Gorgias figure measures: a conversation that advises changes buying behaviour; a conversation that redirects does not.
The special case of products that carry weight: beauty and health
Products that touch skin or health demand a level of rigour a general-purpose chatbot was never built to offer. It is no accident that the major brands — L'Oréal, La Roche-Posay, Vichy — have each developed their own AI skin-assessment tool built into their sites: they identified this need for individual advice well before the current wave of chatbots.
On that ground, two guarantees matter more than anything: a knowledge base fed by the official documentation of the major brands, and a register of guidance reviewed and approved by health professionals — a dermatologist, a paediatrician, a pharmacist. On sensitive cases (pregnancy, allergy, a complaint), the rule has to be fixed: hand over to the retailer's team, without exception. Our page on the AI advisor for beauty and skincare sets out that approach, and reliability, GDPR and responsibility explains how to verify those guarantees before signing.
Frequently asked questions
›Does a chatbot actually increase sales?
It depends entirely on what the chatbot really does. One limited to support (FAQ, order tracking) has no direct effect on sales — that is not its function. A conversation that advises, on the other hand, produces the conversion gap measured above: the advised visitor buys, the other leaves. So the question is not “does a chatbot sell”, but “does this one advise, or redirect?”.
›How long does it take to install an e-commerce chatbot?
For a well-built tool, a few days. On Shopify, from the app authorised inside the store; elsewhere, from a catalogue export, with no access to the retailer's internal systems. A direct Shopify connection can be added afterwards for live stock and orders. Longer timelines usually come from deep technical integration, not from the product advice itself.
›Does a chatbot replace customer service?
No — and be wary of any promise that it does. A good setup handles a large share of requests end to end, but always leaves the wheel to a human team on the cases that need one: a complaint, a sensitive situation, a question outside the catalogue. The team has to be able to take over the conversation at any moment from a dashboard, with every exchange on the record.
›Does a chatbot work on WhatsApp?
Yes, if the tool is built for multiple channels. The right reflex is to check that the site, WhatsApp and social channels share the same memory of the catalogue and of the conversation: a customer who starts on the site and continues on WhatsApp must not start from zero. It is a criterion to verify before signing, not something to assume.
›How do you stop a chatbot inventing answers?
By checking it never recommends outside the retailer's real catalogue, and that it works from an up-to-date knowledge base rather than guesswork. On sensitive subjects (allergy, pregnancy, complaints), the serious guarantee is a fixed rule that hands over to a human, not a vague promise of “reliability”. Always ask how that hand-over is actually triggered.
Alexandre builds Ryma, the AI-powered advisor for online stores, and writes these guides from real deployments, with every figure attributed.